offtracker 2.7.8__zip → 2.10.0__zip

offtracker-2.10.0/offtracker.egg-info/PKG-INFO

@@ -0,0 +1,233 @@
+Metadata-Version: 2.1
+Name: offtracker
+Version: 2.10.0
+Summary: Tracking-seq data analysis
+Home-page: https://github.com/Lan-lab/offtracker
+Author: Runda Xu
+Author-email: xrd18@tsinghua.org.cn
+Requires-Python: >=3.6.0
+Description-Content-Type: text/markdown
+License-File: LICENSE.txt
+
+
+# OFF-TRACKER
+
+OFF-TRACKER is an end to end pipeline of Tracking-seq data analysis for detecting off-target sites of any genome editing tools that generate double-strand breaks (DSBs) or single-strand breaks (SSBs).
+
+## System requirements
+
+* Linux/Unix 
+* Python >= 3.6
+
+## Dependency
+
+```bash
+# We recommend creating a new environment using mamba/conda to avoid compatibility problems
+# If you don't use mamba, just replace the code with conda 
+# Windows systems may not be compatible with pybedtools.
+mamba create -n offtracker -c bioconda blast snakemake pybedtools chromap
+```
+
+
+## Installation 
+
+```bash
+# Activate the environment
+conda activate offtracker
+
+# Direct installation with pip
+pip install offtracker
+
+# (Alternative) Download the offtracker from github
+git clone https://github.com/Lan-lab/offtracker.git 
+cd offtracker
+pip install .
+```
+
+
+## Before analyzing samples
+
+```bash
+# Build blast index (only need once for each genome)
+makeblastdb -input_type fasta -title hg38 -dbtype nucl -parse_seqids \
+-in /Your_Path_To_Reference/hg38_genome.fa \
+-out /Your_Path_To_Reference/hg38_genome.blastdb \
+-logfile /Your_Path_To_Reference/hg38_genome.blastdb.log
+
+# Build chromap index (only need once for each genome)
+chromap -i -r /Your_Path_To_Reference/hg38_genome.fa \
+-o /Your_Path_To_Reference/hg38_genome.chromap.index
+
+# Generate candidate regions by sgRNA sequence (need once for each genome and sgRNA)
+# --name: a user-defined name of the sgRNA, which will be used in the following analysis.
+offtracker_candidates.py -t 8 -g hg38 \
+-r /Your_Path_To_Reference/hg38_genome.fa \
+-b /Your_Path_To_Reference/hg38_genome.blastdb \
+--name 'VEGFA2' --sgrna 'GACCCCCTCCACCCCGCCTC' --pam 'NGG' \
+-o /Your_Path_To_Candidates_Folder
+
+```
+
+
+## Quality control and adapter trimming
+
+```bash
+# Generate snakemake config file for quality control and adapter trimming.
+offtracker_qc.py -t 4 \
+-f /Your_Path_To_Input_Folder \
+--subfolder 0
+
+cd /Your_Path_To_Input_Folder/Trimmed_data
+snakemake -np # dry run to check whether everything is alright
+nohup snakemake --cores 16 1>${outdir}/sm_qc.log 2>&1 &
+
+"""
+Set “--subfolder 0” if the file structure is like: 
+| - Input_Folder
+  | - sample1_R1.fastq.gz
+  | - sample1_R2.fastq.gz
+  | - sample2_R1.fastq.gz
+  | - sample2_R2.fastq.gz
+Set “--subfolder 1” if the file structure is like:
+| - Input_Folder
+  | - Sample1_Folder
+    | - sample1_R1.fastq.gz
+    | - sample1_R2.fastq.gz
+  | - Sample2_Folder
+    | - sample2_R1.fastq.gz
+    | - sample2_R2.fastq.gz
+
+The script “offtracker_qc.py” will create a “Trimmed_data” folder under /Your_Path_To_Input_Folder. 
+If “-o /Your_Path_To_Output” is set, the output will be redirected to /Your_Path_To_Output.
+"""
+```
+
+## Strand-specific mapping of Tracking-seq data 
+
+```bash
+
+# Generate snakemake config file for mapping
+# Results will be generated in /Your_Path_To_Output, if -o is not set, the output will be in the same folder as the fastq files
+offtracker_config.py -t 8 -g hg38 --blacklist hg38 \
+-r /Your_Path_To_Reference/hg38_genome.fa \
+-i /Your_Path_To_Reference/hg38_genome.chromap.index \
+-f /Your_Path_To_Trimmed_Data \
+-o /Your_Path_To_Output \ 
+--subfolder 0 
+
+# Warning: Do not contain "fastq" or "fq" in the folder name, otherwise the program may treat the folder as a fastq file
+# This problem may be fixed in the future
+
+# Run the snakemake program
+cd /Your_Path_To_Fastq
+snakemake -np # dry run
+nohup snakemake --cores 16 1>sm_mapping.log 2>sm_mapping.err &
+
+## about cores
+# --cores of snakemake must be larger than -t of offtracker_config.py
+# parallel number = cores/t
+
+## about output
+# This part will generate "*.fw.scaled.bw" and ".rv.scaled.bw" for IGV visualization
+# "*.fw.bed" and "*.rv.bed" are used in the next part.
+```
+
+
+## Analyzing the genome-wide off-target sites
+
+```bash
+# In this part, multiple samples in the same condition can be analyzed in a single run by pattern recognition of sample names
+
+offtracker_analysis.py -g hg38 --name "VEGFA2" \
+--exp 'Cas9_VEGFA2' \
+--control 'WT' \
+--outname 'Cas9_VEGFA_293' \
+-f /Your_Path_To_Output \
+--seqfolder /Your_Path_To_Candidates
+
+# --name: the same gRNA name you set when running offtracker_candidates.py
+# --exp/--control: add one or multiple patterns of file name in regular expressions
+# If multiple samples meet the pattern, their signals will be averaged. Thus, only samples with the same condition should be included in a single analysis.
+
+# This step will generate Offtracker_result_{outname}.csv
+# Default FDR is 0.05, which can be changed by --fdr. This will empirically make the threshold of Track score around 2.
+# Sites with Track score >=2, which is a empirical threshold, are output regardless of FDR.
+# Intermediate files are saved in ./temp folder, which can be deleted.
+# Keeping the intermediate files can make the analysis faster if involving previously analyzed samples (e.g. using the same control samples for different analyses)
+```
+
+## Off-target sequences visualization
+
+```bash
+# After get the Offtracker_result_{outname}.csv, you can visualize the off-target sites with their genomic sequence with the following command:
+
+offtracker_plot.py --result Your_Offtracker_Result_CSV \
+--sgrna 'GACCCCCTCCACCCCGCCTC' --pam 'NGG'
+
+# The default output is a pdf file with Offtracker_result_{outname}.pdf
+# Assigning a specific output file with another suffix can change the format. e.g., "--output Offtracker_plot.png" will generate a png file.
+# The orange dash line indicates the empirical threshold of Track score = 2
+# Empirically, the off-target sites with Track score < 2 are less likely to be real off-target sites.
+```
+
+
+## Note1, when not using hg38 or mm10
+
+The default setting only includes chr1-chr22, chrX, chrY, and chrM. (only suitable for human and mouse) \
+If you are using reference genomes without "chr" at the beginning, or want to analyze all chromosomes or other species, you can set "--ignore_chr" when running offtracker_config.py to skip chromosome filter.
+
+Currently, this software is only ready-to-use for mm10 and hg38. For any other genome, e.g., hg19, please add a genome size file named "hg19.chrom.sizes" to .\offtracker\utility. Besides, add "--blacklist none" or "--blacklist Your_Blacklist" (e.g., ENCODE blacklist) when running offtracker_config.py, because we only include blacklists for mm10 and hg38.
+
+## Note2
+
+The FDRs in the Tracking-seq result do not reflect the real off-target probability.
+It is strongly recommended to observe the "fw.scaled.bw" and "rv.scaled.bw" using genome browser like IGV to visually inspect each target location from the Tracking-seq result.
+
+
+
+# Example Data
+
+Here are example data that contains reads of chr6 from HEK293T cells edited with Cas9 + sgRNA VEGFA2 and wild type cells:
+
+https://figshare.com/articles/dataset/WT_HEK239T_chr6/25956034
+
+It takes about 5-10 minutes to run the mapping (offtracker_config.py & snakemake) of example data with -t 8 and --cores 16 (2 parallel tasks)
+
+## Signal visualization
+
+After mapping, there will be 4 .bw files in the output folder:
+```bash
+Cas9_VEGFA2_chr6.fw.scaled.bw
+
+Cas9_VEGFA2_chr6.rv.scaled.bw
+
+WT_chr6.fw.scaled.bw
+
+WT_chr6.rv.scaled.bw
+```
+These files can be visualized in genome browser like IGV:
+
+![signal](https://github.com/Lan-lab/offtracker/blob/main/example_output/signals_example.png?raw=true)
+
+The signal (coverage) for each sample is normalized to 1e7/total_reads. As only reads mapping to chr6 were extracted in the example data, the signal range is much higher than that of the whole genome samples. 
+
+## Whole genome off-target analysis
+
+For analyzing the signals (offtracker_analysis.py), it takes about 3-5 minutes and outputs a file named "Offtracker_result_{outname}.csv"
+
+After that, you can visualize the off-target sites with their genomic sequence (offtracker_plot.py) and get an image like this:
+
+![offtarget](https://github.com/Lan-lab/offtracker/blob/main/example_output/sequences_example.png?raw=true)
+
+# Citation
+
+If you use Tracking-seq or OFF-TRACKER in your research, please cite the following paper:
+
+Zhu, M., Xu, R., Yuan, J., Wang, J. et al. Tracking-seq reveals the heterogeneity of off-target effects in CRISPR–Cas9-mediated genome editing. Nat Biotechnol (2024). https://doi.org/10.1038/s41587-024-02307-y
+
+The signal visualization of .bw file here was generated by the Integrative Genomics Viewer (IGV) software. The signal visualization in the Tracking-seq article above was generated by either IGV or pyGenomeTracks:
+
+Robinson, J., Thorvaldsdóttir, H., Winckler, W. et al. Integrative genomics viewer. Nat Biotechnol 29, 24–26 (2011). https://doi.org/10.1038/nbt.1754
+
+Lopez-Delisle L, Rabbani L, Wolff J, Bhardwaj V, Backofen R, Grüning B, Ramírez F, Manke T. pyGenomeTracks: reproducible plots for multivariate genomic data sets. Bioinformatics. 2020 Aug 3:btaa692. doi: 10.1093/bioinformatics/btaa692.
+

offtracker-2.10.0/offtracker.egg-info/SOURCES.txt

@@ -0,0 +1,28 @@
+LICENSE.txt
+MANIFEST.in
+README.md
+setup.py
+offtracker/X_offplot.py
+offtracker/X_offtracker.py
+offtracker/X_sequence.py
+offtracker/__init__.py
+offtracker/_version.py
+offtracker.egg-info/PKG-INFO
+offtracker.egg-info/SOURCES.txt
+offtracker.egg-info/dependency_links.txt
+offtracker.egg-info/requires.txt
+offtracker.egg-info/top_level.txt
+offtracker/snakefile/Snakefile_QC.smk
+offtracker/snakefile/Snakefile_offtracker.smk
+offtracker/utility/1.1_bed2fr.py
+offtracker/utility/1.3_bdg_normalize_v4.0.py
+offtracker/utility/bedGraphToBigWig
+offtracker/utility/hg38.chrom.sizes
+offtracker/utility/mm10.chrom.sizes
+offtracker/utility/offtracker_blacklist_hg38.merged.bed
+offtracker/utility/offtracker_blacklist_mm10.merged.bed
+scripts/offtracker_analysis.py
+scripts/offtracker_candidates.py
+scripts/offtracker_config.py
+scripts/offtracker_plot.py
+scripts/offtracker_qc.py

{offtracker-2.7.8 → offtracker-2.10.0}/scripts/offtracker_analysis.py

@@ -26,6 +26,8 @@ def main():
     parser.add_argument('--name'         , type=str, required=True,    help='custom name of the sgRNA' )
     parser.add_argument('--exp'          , type=str, default='all',    nargs='+', help='A substring mark in the name of experimental samples. The default is to use all samples other than control' )
     parser.add_argument('--control'      , type=str, default='none',   nargs='+', help='A substring mark in the name of control samples. The default is no control. "others" for all samples other than --exp.' )
+    parser.add_argument('--fdr'          , type=int, default=0.05,     help='FDR threshold for the final result. Default is 0.05.')
+    parser.add_argument('--score'        , type=int, default=2,        help='Track score threshold for the final result. Default is 2.')
     parser.add_argument('--smooth'       , type=int, default=1,        help='Smooth strength for the signal.')
     parser.add_argument('--window'       , type=int, default=3,        help='Window size for smoothing the signal.')
     parser.add_argument('--binsize'      , type=int, default=100,      help='Window size for smoothing the signal.')
@@ -41,6 +43,7 @@ def main():
     parser.add_argument('--overwrite'    , action='store_true', help='Whether to overwrite existed dataframes.' )
     parser.add_argument('--clean'        , action='store_true', help='Whether to remove temp files')
 
+
     args = parser.parse_args()
 
     print(f'Runing offtracker verision: {offtracker.__version__}')
@@ -49,6 +52,8 @@ def main():
     sgRNA_name = args.name
     pattern_exp = args.exp
     pattern_ctr = args.control
+    fdr_thresh = args.fdr
+    score_thresh = args.score
     binsize = args.binsize
     flank_max = args.flank_max
     flank_regions = args.flank_regions
@@ -93,6 +98,8 @@ def main():
         all_sample_files.extend( bdg_files )
     all_sample_files = pd.Series(all_sample_files)
     all_sample_names = pd.Series(all_sample_names)
+    print('all sample names in the folders:')
+    print(all_sample_names)
     print('your string pattern for experimental groups: ', pattern_exp)
     ctr_samples = []
     if pattern_ctr == 'none':
@@ -155,8 +162,11 @@ def main():
         df_bdg.columns = ['chr','start','end','residual']
         # 将 df_bdg 按照染色体分组
         sample_groups = df_bdg.groupby('chr')
+        # 2024.06.03. fix a bug that df_bdg has less chr than df_candidate
+        total_chr = df_bdg['chr'].unique()
+        df_candidate_sub_temp = df_candidate_sub[df_candidate_sub['chr'].isin(total_chr)]
         # 将 df_candidate_sub 按照染色体分组
-        candidate_groups = df_candidate_sub.groupby('chr')
+        candidate_groups = df_candidate_sub_temp.groupby('chr')
 
         # 定义一个空的列表，用于存储每个染色体的数据
         chrom_list = []
@@ -234,7 +244,8 @@ def main():
             df_score = pd.concat([df_score, df_exp, df_ctr], axis=1)
         else:
             df_score = pd.concat([df_score, df_exp], axis=1)
-        df_score = df_score.copy()
+        # 2024.06.03. 跑样例数据时，只有一个 chr6, 其他都是 nan, 不删除会导致后续计算出错
+        df_score = df_score.dropna().copy()
         df_score.to_csv(output)
     
     ##########################
@@ -335,12 +346,16 @@ def main():
         print('mean_score:{:.3f};std:{:.3f}'.format(mu,std))
         # pv and fdr
         df_result['pv'] = df_result[f'log2_track_score'].apply( lambda x: norm.sf(x,loc=mu,scale=std) )
-        df_result['pv'].clip(lower=1e-320,inplace=True)
+        df_result['pv'] = df_result['pv'].clip(lower=1e-320)
         df_result['fdr'] = offtracker.fdr(df_result['pv'])
         df_result['rank'] = range(1,len(df_result)+1)
         df_result.to_csv(output)        
-
-        df_output = df_result[df_result['fdr']<=0.05].copy()
+        # 2024.06.03. 以防 fdr<=fdr_thresh 滤掉了 track_score>=2 的位点
+        bool_fdr = df_result['fdr']<=fdr_thresh
+        bool_score = df_result['track_score']>=score_thresh
+        # 2025.06.05. BE可能会形成单边信号，导致 track_score 为负数，也保留
+        bool_neg_score = df_result['track_score']<0
+        df_output = df_result[bool_fdr|bool_score|bool_neg_score].copy()
         if pattern_ctr != 'none':
             df_output = df_output[['target_location', 'best_strand','best_target','deletion','insertion','mismatch',
                                    'exp_L_length', 'exp_R_length','ctr_L_length','ctr_R_length','L_length','R_length','signal_length',

offtracker-2.10.0/scripts/offtracker_candidates.py

@@ -0,0 +1,318 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+# 2023.10.27. v2.0: 2.0以target_location midpoint为中心，因此取消 pct 计算
+# 2023.12.06. v2.1: 2.1增加 cleavage_site 推测, 修正 deletion 错位, 以 cleavage_site 为中心
+# 2025.04.25. 修正大小写问题
+# 2025.06.11. 调整跳过已存在的candidates的代码顺序
+
+import os,sys,re,time
+from itertools import product, permutations
+
+if sys.version_info < (3,0):
+    import platform
+    raise Exception(f'python3 is needed, while running {platform.python_version()} now')
+
+import offtracker
+import offtracker.X_sequence as xseq
+script_dir = os.path.abspath(os.path.dirname(offtracker.__file__))
+script_folder= os.path.join(script_dir, 'mapping')
+
+import argparse
+import pandas as pd
+import pybedtools
+import multiprocessing as mp
+from Bio.Blast.Applications import NcbiblastnCommandline 
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.description='Generate candidate regions by sgRNA sequence'
+    parser.add_argument('--sgrna' ,       type=str, required=True, help='One sgRNA sequence without PAM' )
+    parser.add_argument('--pam'   ,       type=str, required=True, help='The protospacer adjacent motif' )
+    parser.add_argument('--pam_location', type=str, default='downstream', help='Upstream or downstream, default is downstream (Cas9)' )
+    parser.add_argument('--name'  ,       type=str, required=True, help='custom name of the sgRNA' )
+    parser.add_argument('-r','--ref'    , type=str, required=True, help='The fasta file of reference genome')
+    parser.add_argument('-b','--blastdb', type=str, required=True, help='blast database')
+    parser.add_argument('-o','--outdir' , type=str, required=True, help='The output folder')
+    parser.add_argument('-g','--genome' , type=str, default='hg38', help='File of chromosome sizes, or "hg38", "mm10" ')
+    parser.add_argument('-t','--thread' , type=int, default=4,     help='Number of threads for parallel computing')
+    parser.add_argument('--quick_mode'  , action='store_true',  help='BLAST faster but less candidates.')
+
+    args = parser.parse_args()
+    
+    
+    if (args.genome == 'hg38') or (args.genome == 'mm10'):
+        dir_chrom_sizes = os.path.join(script_folder, f'{args.genome}.chrom.sizes')
+    else:
+        dir_chrom_sizes = args.genome
+    
+    sgRNA_name = args.name
+    sgRNA_seq  = args.sgrna
+    PAM = args.pam
+    PAM_loc = args.pam_location.lower()
+    n_threads  = args.thread
+    dir_output = args.outdir
+    if not os.path.exists(dir_output):
+        os.makedirs(dir_output)
+    dir_ref_fa = args.ref
+    blast_db   = args.blastdb
+    quick_mode = args.quick_mode
+    
+    # parameters for alignment
+    half_width = 100
+    pct_params = 1.0
+    frag_len= half_width*2
+    dir_df_candidate = os.path.join(dir_output, f'df_candidate_{sgRNA_name}.csv')
+    if os.path.isfile(dir_df_candidate):
+        print(f'{dir_df_candidate} exists, skipped.') 
+        return 'skipped'
+    
+    sgRNA_seq = sgRNA_seq.upper()
+    PAM = PAM.upper()
+    dir_sgRNA_fasta = os.path.join(dir_output, f'{sgRNA_name}_PAM.fasta')
+    dir_sgRNA_blast = os.path.join(dir_output, f'{sgRNA_name}_PAM.blast')
+    dir_sgRNA_bed   = os.path.join(dir_output, f'{sgRNA_name}_PAM.bed')
+    
+    if PAM_loc == 'downstream':
+        possible_sgRNA_PAM = list(product([sgRNA_seq],xseq.possible_seq(PAM)))
+    elif PAM_loc == 'upstream':
+        possible_sgRNA_PAM = list(product(xseq.possible_seq(PAM),[sgRNA_seq]))
+    else:
+        raise Exception(f'PAM_location should be "upstream" or "downstream", while {PAM_loc} is given.')
+    possible_sgRNA_PAM = [''.join(combination) for combination in possible_sgRNA_PAM]
+    n_seq = len(possible_sgRNA_PAM)
+    
+    ID = pd.Series(['seq']*n_seq) + pd.Series(range(1,n_seq+1)).astype(str)
+    df_sgRNA_PAM = pd.DataFrame({'ID':ID,'sequence':possible_sgRNA_PAM})
+    xseq.write_fasta(df_sgRNA_PAM, dir_sgRNA_fasta)
+    
+    #########
+    # BLAST #
+    #########
+    if os.path.isfile(dir_sgRNA_blast):
+        print(f'{dir_sgRNA_blast} exists, skipped.')
+    else:
+        if quick_mode:
+            print('Using quick mode for BLAST')
+            blastx_cline = NcbiblastnCommandline(query=dir_sgRNA_fasta, task='blastn-short',out=dir_sgRNA_blast,
+                                                db=blast_db, evalue=10000,outfmt=6, num_threads=n_threads,
+                                                gapopen=4, gapextend=2, reward=2, word_size=5, dust='no', soft_masking=False)
+        else:
+            blastx_cline = NcbiblastnCommandline(query=dir_sgRNA_fasta, task='blastn-short',out=dir_sgRNA_blast,
+                                                db=blast_db, evalue=10000,outfmt=6, num_threads=n_threads,
+                                                gapopen=4, gapextend=2, reward=2, word_size=4, dust='no', soft_masking=False)   
+        print(f'BLAST for candidate off-target sites of {sgRNA_name}.')
+        blastx_cline()
+        print(f'BLAST finished.')
+    
+    ##############
+    # Output bed #
+    ##############
+    
+    blast_regions = pd.read_csv(dir_sgRNA_blast, sep='\t',header=None)
+    blast_regions.columns = ['query acc.','chr','% identity','alignment length','mismatches','gap opens','q. start','q. end','st','ed','evalue','bit score']
+    blast_regions = blast_regions[blast_regions.evalue<10000]
+    
+    # reverse strand 
+    blast_regions['reverse'] = (blast_regions['st']>blast_regions['ed']).astype(int)
+    blast_regions_f = blast_regions[blast_regions.reverse==0].copy()
+    blast_regions_r = blast_regions[blast_regions.reverse==1].copy()
+    temp = blast_regions_r['st'].copy()
+    blast_regions_r['st'] = blast_regions_r['ed']
+    blast_regions_r['ed'] = temp
+    blast_regions = pd.concat([blast_regions_f, blast_regions_r])
+    # sort and add location
+    blast_regions = blast_regions.sort_values('evalue').reset_index(drop=True)
+    blast_regions['location']=blast_regions['chr'].str[:] + ':' + blast_regions['st'].astype(str).str[:] + '-' + blast_regions['ed'].astype(str).str[:]
+    blast_regions = blast_regions.drop_duplicates(subset='location').copy()
+    
+    # alignment length 筛选
+    len_sgRNA=len(sgRNA_seq)
+    min_len = len_sgRNA-8
+    blast_regions = blast_regions[blast_regions['alignment length']>=min_len].copy().reset_index(drop=True)
+    blast_regions = blast_regions.reindex(columns = ['chr', 'st', 'ed' , 'query acc.', '% identity', 'alignment length', 'mismatches',
+        'gap opens', 'q. start', 'q. end', 'evalue', 'bit score', 'reverse', 'location'] )
+    
+    # 输出 bed 用于后续 alignment score 计算
+    blast_regions_bed = blast_regions[['chr','st','ed']]
+    xseq.write_bed(blast_regions_bed, dir_sgRNA_bed)
+    # 对 bed 进行排序但不合并
+    a = pybedtools.BedTool(dir_sgRNA_bed)
+    a.sort(g=dir_chrom_sizes).saveas( dir_sgRNA_bed )
+    print(f'Output {sgRNA_name}_PAM.bed')
+    
+    
+    ###################
+    # alignment score #
+    ###################
+
+    #########
+    # 读取 blast bed
+    #########
+    common_chr = pd.Series(['chr']*23).str[:] + pd.Series(range(23)).astype(str).str[:]
+    common_chr = pd.concat([common_chr, pd.Series(['chrX','chrY'])]).to_numpy()
+
+    bed_short = xseq.X_readbed(dir_sgRNA_bed)
+    bed_short = bed_short[bed_short['chr'].isin(common_chr)].copy()
+    bed_short['midpoint'] = ((bed_short['st'] + bed_short['ed'])/2).astype(int)
+    bed_short['st'] = bed_short['midpoint'] - half_width 
+    bed_short['ed'] = bed_short['midpoint'] + half_width
+    bed_short.loc[bed_short['st']<0,'st']=0
+    bed_short = bed_short.drop_duplicates()        
+
+    #########
+    # 根据 bed_f 位点 ed 前后 half_width 取基因组序列
+    #########
+
+    temp_bed = os.path.join(dir_output, 'temp.bed')
+    xseq.write_bed(bed_short.iloc[:,:3], temp_bed)
+    a = pybedtools.BedTool(temp_bed)
+    fasta = pybedtools.example_filename(dir_ref_fa)
+    a = a.sequence(fi=fasta)
+    with open(a.seqfn) as f:
+        fasta = {} 
+        for line in f:
+            line = line.strip() # 去除末尾换行符
+            if line[0] == '>':
+                header = line[1:]
+            else:
+                sequence = line
+                fasta[header] = fasta.get(header,'') + sequence
+
+    # pybedtools 得到位置 chrA:X-Y 时，X数字会往左多1bp
+
+    #########
+    # local alignment
+    #########
+    # 生成 DNA_matrix
+    mismatch_score = 0.01
+    base_codes = list(xseq.ambiguous_nt.keys())
+    all_base_pairs = list(permutations(base_codes,2)) + [(x,x) for x in base_codes]
+    DNA_matrix = {x : xseq.get_base_score(*x, mismatch_score=mismatch_score) for x in all_base_pairs}
+    # 添加 PAM
+    if PAM_loc == 'downstream':
+        sgRNA_PAM_fw = sgRNA_seq + PAM
+    else:
+        sgRNA_PAM_fw = PAM + sgRNA_seq
+
+    sgRNA_PAM_rv = xseq.reverse_complement(sgRNA_PAM_fw)
+
+    list_args_fw=[]
+    list_args_rv=[]
+    for a_key, a_seq in fasta.items():
+        # 2025.04.25 修正大小写问题
+        a_seq = re.sub('[^ATCG]','N',a_seq.upper())
+        list_args_fw.append( [a_key, sgRNA_PAM_fw, a_seq, frag_len, DNA_matrix, mismatch_score] )
+        list_args_rv.append( [a_key, sgRNA_PAM_rv, a_seq, frag_len, DNA_matrix, mismatch_score] )
+    st = time.time()
+    with mp.Pool(n_threads) as p:
+        list_align_forward = p.starmap(xseq.sgRNA_alignment, list_args_fw)
+    ed = time.time()
+    print('align_forward:{:.2f}'.format(ed-st))
+    st = time.time()
+    with mp.Pool(n_threads) as p:
+        list_align_reverse = p.starmap(xseq.sgRNA_alignment, list_args_rv)
+    ed = time.time()
+    print('align_reverse:{:.2f}'.format(ed-st))
+    #
+    df_align_forward = pd.DataFrame(list_align_forward, columns= ['fw_score','fw_pct','fw_target','fw_location','fw_deletion','fw_insertion','fw_mismatch'])
+    df_align_reverse = pd.DataFrame(list_align_reverse, columns= ['rv_score','rv_pct','rv_target','rv_location','rv_deletion','rv_insertion','rv_mismatch'])
+    df_align_reverse['rv_target'] = df_align_reverse['rv_target'].apply(xseq.reverse_complement)
+    df_candidate = pd.concat([df_align_forward,df_align_reverse],axis=1)
+    df_candidate['location'] = fasta.keys()
+    df_candidate['alignment_score'] = df_candidate[['fw_score','rv_score']].max(axis=1)
+    #df_candidate['fw_score_2'] = df_candidate['fw_score']*(pct_params-df_candidate['fw_pct'].abs())
+    #df_candidate['rv_score_2'] = df_candidate['rv_score']*(pct_params-df_candidate['rv_pct'].abs())
+    #df_candidate['best_seq_score'] = df_candidate[['fw_score_2', 'rv_score_2']].max(axis=1)
+    #df_candidate['best_strand'] = df_candidate[['fw_score_2', 'rv_score_2']].idxmax(axis='columns').replace({'fw_score_2':'+', 'rv_score_2':'-'})
+    #df_candidate.loc[df_candidate['fw_score_2']==df_candidate['rv_score_2'],'best_strand']='equal_score'
+    df_candidate['best_seq_score'] = df_candidate[['fw_score', 'rv_score']].max(axis=1)
+    df_candidate['best_strand'] = df_candidate[['fw_score', 'rv_score']].idxmax(axis='columns').replace({'fw_score':'+', 'rv_score':'-'})
+    df_candidate.loc[df_candidate['fw_score']==df_candidate['rv_score'],'best_strand']='equal_score'
+            
+    # GG check
+    # 2023.12.05 增加 cleavage_site 推测
+    list_best_target = []
+    list_best_location = []
+    list_cleavage_site = []
+    list_delete = []
+    list_insert = []
+    list_mismat = []
+    list_GG = []
+    for a_row in df_candidate.iterrows():
+        if a_row[1]['best_strand']=='+':
+            list_best_target.append(a_row[1]['fw_target'])
+            list_best_location.append(a_row[1]['fw_location'])
+            list_cleavage_site.append(int(a_row[1]['fw_location'].split('-')[1]) - 6)
+            list_delete.append(a_row[1]['fw_deletion'])
+            list_insert.append(a_row[1]['fw_insertion'])
+            list_mismat.append(a_row[1]['fw_mismatch'])
+            if a_row[1]['fw_target'][-2:]=='GG':
+                list_GG.append('OK')
+            else:
+                list_GG.append('NO')                     
+        elif a_row[1]['best_strand']=='-':
+            list_best_target.append(a_row[1]['rv_target'])
+            list_best_location.append(a_row[1]['rv_location'])
+            list_cleavage_site.append(int(a_row[1]['rv_location'].split('-')[0].split(':')[1]) + 5)
+            list_delete.append(a_row[1]['rv_deletion'])
+            list_insert.append(a_row[1]['rv_insertion'])
+            list_mismat.append(a_row[1]['rv_mismatch'])
+            if a_row[1]['rv_target'][-2:]=='GG':
+                list_GG.append('OK')
+            else:
+                list_GG.append('NO')  
+        else:
+            if a_row[1]['fw_target'][-2:]=='GG':
+                list_best_target.append(a_row[1]['fw_target'])
+                list_best_location.append(a_row[1]['fw_location'])
+                list_cleavage_site.append(int(a_row[1]['fw_location'].split('-')[1]) - 6)
+                list_delete.append(a_row[1]['fw_deletion'])
+                list_insert.append(a_row[1]['fw_insertion'])
+                list_mismat.append(a_row[1]['fw_mismatch'])
+                list_GG.append('OK_same_score')
+            # 发现没有 GG 则看 RC
+            elif a_row[1]['rv_target'][-2:]=='GG':
+                list_best_target.append(a_row[1]['rv_target'])
+                list_best_location.append(a_row[1]['rv_location'])
+                list_cleavage_site.append(int(a_row[1]['rv_location'].split('-')[0].split(':')[1]) + 5)
+                list_delete.append(a_row[1]['rv_deletion'])
+                list_insert.append(a_row[1]['rv_insertion'])
+                list_mismat.append(a_row[1]['rv_mismatch'])
+                list_GG.append('OK_same_score')
+            else:
+                list_best_target.append(a_row[1]['fw_target'])
+                list_best_location.append(a_row[1]['fw_location'])
+                list_cleavage_site.append(int(a_row[1]['fw_location'].split('-')[1]) - 6)
+                list_delete.append(a_row[1]['fw_deletion'])
+                list_insert.append(a_row[1]['fw_insertion'])
+                list_mismat.append(a_row[1]['fw_mismatch'])                    
+                list_GG.append('NO_same_score')
+    # 记入 df_candidate
+    df_candidate['deletion'] = list_delete
+    df_candidate['insertion'] = list_insert
+    df_candidate['mismatch'] = list_mismat
+    df_candidate['GG'] = list_GG
+    df_candidate['best_target'] = list_best_target
+    df_candidate['target_location'] = list_best_location
+    df_candidate['cleavage_site'] = list_cleavage_site
+
+    # 2.0 更新一下格式
+    df_candidate = df_candidate.drop_duplicates(subset=['target_location']).reset_index(drop=True)
+    df_candidate = pd.concat([xseq.bedfmt(df_candidate['target_location']), df_candidate],axis=1)
+    # df_candidate['midpoint'] = ((df_candidate['ed'] + df_candidate['st'])/2).astype(int)
+    df_candidate = xseq.add_ID(df_candidate, midpoint='cleavage_site')
+
+    df_candidate.to_csv(dir_df_candidate)
+    print(f'Output df_candidate_{sgRNA_name}.csv')
+    os.remove(temp_bed)
+
+    return 'Done!'
+    
+
+if __name__ == '__main__' :
+    result = main()
+    print(result)
+
+
+